dangole has quit [Remote host closed the connection]
Tapper1 has joined #openwrt-devel
svanheule_ has joined #openwrt-devel
Tapper1 has quit []
Tapper has quit [Ping timeout: 480 seconds]
svanheule has quit [Ping timeout: 480 seconds]
cmonroe has joined #openwrt-devel
cmonroe_ has quit [Ping timeout: 480 seconds]
cmonroe has quit [Ping timeout: 480 seconds]
danitool has quit [Ping timeout: 480 seconds]
amaumene has joined #openwrt-devel
<amaumene>
hi guys, I've opened a task here https://bugs.openwrt.org/index.php?do=details&task_id=4239&only_watched=1 it seems we are missing patches from 5.13 to have bridges in the list of devices when enabling hardware offload with nft. what's weird is it's working fine with software offload, so I'm not 100% sure this is not an issue with nft itself.
<amaumene>
if anyone can help to confirm/investigate I'm happy to spend more time on this :)
<aparcar>
Grommish: is there a working rust setup now?
<Grommish>
neggles: Your build has no ethernet forme
<Grommish>
neggles: even with the reset /etc/config/network
<neggles>
interesting
<neggles>
Grommish: nothing under /sys/class/net?
<Grommish>
aparcar: Depends on what you mean by working. Right now, the only thing I'm testing is mips64 because I don't have any other devices.. In theory, it should work though I'm still tryiing to get things settled as far as where things go
<Grommish>
neggles: Yep
<neggles>
Grommish: hmm. odd. possibly kernel config is missing some bits
<neggles>
shouldn't be... all i did was flip on some nand stuff
<neggles>
i also uh, i assume you didn't select failsafe mode right? cause failsafe's network config override stuff would stop /etc/config/network applying
<Grommish>
No failsafe
<neggles>
ok good
<Grommish>
I'm not showing dnsmasq
<Grommish>
Which is why I don't have DNS or dhcp ;p
<neggles>
dnsmasq is almost certainly not present in this image but dhcpcd should be
<neggles>
alright alright lemme just rebuild this without half the packages cut out (i was trying to make a tiny kernel i could fit lzma'd in NOR)
<Grommish>
Wait.. I started this chasing down a mem leak I attributed to dnsmasq and/or adblock
<Grommish>
You don't use dnsmasq, but I do and I get stintel does
<neggles>
yup no dnsmasq in here
<Grommish>
err I'd bet stintel does I mean
<neggles>
and... if dnsmasq was causing trouble, that might explain why iperf didn't do anything
<Grommish>
I found a dnsmasq memleak reported in like ubuntu or something recently
<Grommish>
which is why I went looking
<neggles>
shall i rebuild with dnsmasq?
<stintel>
when the ERL was still my backup router, dnsmasq would not run unless it became primary
<Grommish>
Sure, if you don't mind tainting your build ;D
<stintel>
so I'm pretty sure dnsmasq is not the problem
<neggles>
meh, ./scripts/env save :P
<Grommish>
I'm at 39420 (nice) on RAM used
rua has joined #openwrt-devel
<neggles>
quick, write 30mib to /tmp
<Grommish>
and have been around there since boot
<Grommish>
I just tumepd 13mb x3 to tmp
<neggles>
heh, i was aiming for hitting 69420 used kib :P
<Grommish>
5.4 is the current Octeon target revision
<ynezz>
ah, ok, so it's not related to dnsmasq
<Grommish>
5.10 has memleak
<stintel>
ynezz: octeon has a severe memory leak with 5.10, so we decided in the last meeting we will not bump it because my backup router died after ~8h, it's not acceptable
<Grommish>
ynezz: Dunno yet that the thing.. When neggles just turns on dnsmasq, the leak came back.. it was no there prior to it being included
<stintel>
the alternative is: we drop octeon :)
<stintel>
so someone is going to have to figure out that memleak, or no octeon in next release
<Grommish>
ynezz: I run dnsmasq by rote, he doesn't and he never saw the leak
<Grommish>
only stintel and I did
<stintel>
maintainers response was "kmemleak false positives"
<stintel>
very useful
<Grommish>
and very "not my problem" hehe
<neggles>
ohhhhhhhhhh yeah
<neggles>
it's ipv6
<Grommish>
Easy enough for me to test
<neggles>
i just jumped 6mib of RAM usage in the space of 20 seconds
<neggles>
iperf3 ipv6
<stintel>
neggles: don't come to conclusions too quickly though
<stintel>
RAM usage fluctuates weirdly during several iperf runs
<neggles>
i find it levels out and the highest it gets to is largely dependent on peak b/w
<neggles>
but, you have a point
<stintel>
I remember seeing similar things when I was testing multicast iperf ;)
<Grommish>
I'm building out a no v6 build because I have no shame apparently
<ynezz>
did anyone tried to bisect it?
svanheule_ has quit [Quit: svanheule_]
svanheule has joined #openwrt-devel
<Grommish>
ynezz: When I asked about bisecting the Openwrt kernel.. I was laughed at
<neggles>
well there's also the absolutely horrifying hackjob that is "octeon, but kernel 5.10"
<neggles>
because marvell won't give us the SDK version with 5.4 in it
<neggles>
so we're dragging up patches from 4.9 or 4.14...
<stintel>
ynezz: bisecting kernel 5.4 to 5.10 on OpenWrt ... please advise how to pull that off :)
<stintel>
using external git kernel excludes our patches and then the thing wouldn't boot\
<Grommish>
and apparantly they tried to pull the driver out and then shoved it back in at the last miniute
<neggles>
sorta
<neggles>
gregkh, being gregkh, decided to yeet it because nobody had touched it in a while
<neggles>
(and it's a mess) - then marvell devs went "hey no don't do that! we're gonna fix it, we swear!"
<stintel>
nbd: question, when classifying traffic with qosify, should the outgoing packet (e.g. on WAN) have the configured DSCP value set, or is that only used internally as ISP most likely ignores that anyway?
<neggles>
it is quite frustrating that Marvell seem to be doing quite well with upstreaming Armada, OcteonTX, and their switch ASIC stuff, but they won't even just *publish the octeon SDK*
<neggles>
ok, I may have unfairly blamed ipv6
<stintel>
neggles: octeon too old :)
<stintel>
same shit with wifi5
<neggles>
except it's not! they have a MIPS octeon that only came out in 2020!
<ynezz>
stintel: well, you need to first strip that patchset down to bare minimum, boot either from RAM/NFS etc.
<stintel>
and soon with wifi6 as wifi7 is being worked on
<Grommish>
I'm just glad the people who know what they are actually doing are equally as confused BTW
<nbd>
stintel: outgoing packets should have the dscp value set
<stintel>
nbd: ok, thanks. then we need to debug why they haven't
<neggles>
Grommish: haha, do not make the mistake of assuming I know what I'm doing :P
<Grommish>
neggles: I laid no blame ;p
<stintel>
ynezz: definitely not easy to pull off, going to take a fair amount of work, and it's not going to be for everyone most likely
<stintel>
and as I retired my ERL as backup router, it's low prio for me
<Grommish>
ynezz: Honeslty, I think isolating the commits was the part that wasn't known how to do :) At least for me
<stintel>
I think it's more likely to be a change in network subsystem that requires a modification in the driver that was overlooked
<ynezz>
yeah, could be
<neggles>
definitely not ipv6
<neggles>
told dnsmasq to only bind to the v4 addr
<neggles>
still tickin' upwards
<Grommish>
So, we are back to why your builds don't have it
<Grommish>
but main does
<stintel>
I don't even have dnsmasq in my snic10 .config
<neggles>
but do you have device packages
<stintel>
and I saw increasing memory while not routing
<neggles>
i turned that off
<f00b4r0>
jow: I'm preparing a PR for coova-chilli as discussed yesterday, however I'm stuck with the "wan" service interface trigger: until the upstream interface is up and the default route is too, there is no way to programmatically detect which interface is the actual wan interface, is there? So when chilli init is processed, if the network isn't up when the trigger registers, if I use network_find_wan I will get no result?
<stintel>
"device packages"?
<neggles>
'select all device specific packages by default'
<stintel>
no
<neggles>
and i made exactly one change between this image (which has the problem, whenever dnsmasq is running) and the previous image (which does not have the problem)
<neggles>
changed dnsmasq from m to y
<neggles>
hmm i am not sure that it *is* still ticking up actually, dnsmasq running, bound to v4 only
<neggles>
will give it a minute or five
<ynezz>
what about comparing perf output on working and not working kernel? if there are huge allocations differencies then it should be visible
<neggles>
oh yeah there it goes
<aparcar>
Grommish: what is your opinion on downloading a pre build rust compiler? I'm asking since we also offer to download a pre build bpf thing which reduces compile time by an hour or something
<Grommish>
aparcar: Preferrable. People had issues with using something like rustup (which turns out couldn't use anyway) because it was pre-compiled, but if the build system is building the dist archives, I can't see why they'd complain.. it currently builds a dist archive and uses that to install anyway
<Grommish>
I already split the host and target into different archives
<Grommish>
aparcar: I mean.. ideally, we'd offer a full LLVM, not just for bpf
<Grommish>
aparcar: There isn't a reason not to at that point.. rust builds out llvm, libc, clang
<Grommish>
There were questions about using the bpf llvm, but it was suggested to better keep them separate
<robimarko>
f00b4r0 is the author, so he can share more details
<rmilecki>
thanks
<robimarko>
I know that there has been an effort to fix the partial erase and write in an upstream friendly manner
<robimarko>
But I dont know whats the status
<dlg>
robimarko: are you in the mood to talk about uboot and ipq4019s?
<robimarko>
Sure
<f00b4r0>
rmilecki: partial erase is completely fsckd up yes. #3271 is the correct fix, unfortunately it's not moving.
Guest754 has quit [Ping timeout: 480 seconds]
<rmilecki>
f00b4r0: ah, 4K erase sounds nice
<neggles>
Grommish / stintel: hm. dnsmasq still running, no memory usage creep, but it bumped up when i did a dns lookup. idk, something's screwy. not sure how to work out what.
<robimarko>
dlg: What problem are you having with u-boot on IPQ4?
<f00b4r0>
rmilecki: if you can help this PR get some traction, it would be very helpful. There are other affected targets iirc. Basically any target that sets 4K_SECTORS + 4K_LIMIT and expects larger partitions to use 64K EB will see 4K EB being used, which can mess sysupgrade due to EB mismatch
<dlg>
robimarko: the short version is i dont know enough
<rmilecki>
yeah, that needs a big cleanup, all that 4B and 64B mess
<dlg>
robimarko: the long version is i got an rb450gx4 so i could work toward hacking on an ethernet switch phy in openbsd, and didnt realise how much of a gap there is between turning the board on and loading my kernel
<f00b4r0>
mikrotik devices "can" live without 4K, that just means bootloader configuration can no longer happen in OS. It's a hindrance, but it's not lethal. However I think the variable EB feature is nevertheless desirable
<robimarko>
dlg: Well, RB450Gx4 doesnt even use U-boot
Tapper has joined #openwrt-devel
<robimarko>
They use MikroTiks Routerboot
<robimarko>
Which is way more limited
<robimarko>
And we essentially hack around its limtits all the time
<dlg>
robimarko: i know that now. but i have routerboot loading mainline u-boot
<dlg>
and im basically where i would be if i was booting u-boot from metal
<dlg>
my problem is getting the sdhci controller working now
<robimarko>
Well not really
<robimarko>
I wouldnt suggest doing that
<dlg>
which could be because of missing clocks and regulator support
<robimarko>
If you want U-boot replace RouterBoot with it
<robimarko>
And yeah, the SDHCI controller doesnt work
<robimarko>
At least not when I tried it last time
<robimarko>
Clocks are missing as well as regulator support
<robimarko>
But I think even with the clocks it wasnt working
<robimarko>
I basically added just bare minimum to boot from SPI-NOR or SPI-NAND on IPQ4019
<robimarko>
But I can tell you that ethernet support is in the works
<dlg>
cool
<dlg>
but sdhci would still suck if i was on something like a habanero, right?
<robimarko>
Why?
<robimarko>
Its got a 32MB of NOR
<robimarko>
So you can boot whatever kernel you want
<dlg>
unless the kernel is on the sdcard :(
<robimarko>
Well yeah
<robimarko>
Unfortunately, I didnt have a need for it so I only tried the easy way
<robimarko>
To hardcode the clock and regulator and thats it
<karlp>
(is the plan touse sdcard long term anyway? can you not just keep booting over network for testing bsd as well?)
<dlg>
either would work
<dlg>
both would be great
<dlg>
robimarko: do you have a diff from when you tried that?
<robimarko>
Unfortunately not
<robimarko>
Its been a while
<dlg>
fair enough
<robimarko>
Over a year
<robimarko>
I mean, if you dont care you can use Qualcomms U-boot
<robimarko>
Its public
<robimarko>
And everything works there, but there are limits as they have hacked it like always
<robimarko>
GL-inet probably just forked the original
<dlg>
cool, i'll have a look
<dlg>
thank you
<robimarko>
I have pretty much destroyed my Habanero DVK
<robimarko>
With mods so I dont really use it anymore
<dlg>
"maybe a hammer will fix sdhci"?
<robimarko>
Unfortunatelly not
robimarko has quit [Remote host closed the connection]
SamantazFox_ has quit [Ping timeout: 480 seconds]
<rmilecki>
ah, i found it, MTD_ERASE_PARTIAL doesn't work with NAND
<rmilecki>
NAND has ._write_oob pointer set instead of ._write
<rmilecki>
so doing part->parent->_write() crashes kernel
Misanthropos has quit [Read error: Connection reset by peer]
minimal has joined #openwrt-devel
mattytap_ has joined #openwrt-devel
mattytap has quit [Ping timeout: 480 seconds]
Misanthropos has joined #openwrt-devel
pmelange has joined #openwrt-devel
pmelange has left #openwrt-devel [#openwrt-devel]
danitool has joined #openwrt-devel
SamantazFox_ has joined #openwrt-devel
valku has joined #openwrt-devel
<stintel>
nbd: is there any documentation on how the dns based classification works? it doesn't seem to work, and udnssnoop is undocumented
madwoota has quit [Read error: Connection reset by peer]
madwoota has joined #openwrt-devel
<stintel>
I also don't see the add_dns_host call in the dnsmasq patch
<stintel>
and is there an ubus call to show whatever has been added by add_dns_host ?
<stintel>
I don't see anything in dump or status
<rsalvaterra>
Heads up: don't try Linux 5.10.94 on MT7622 unless you have a serial cable available. Something seems to have broken upstream. :/
<aparcar>
anyone facing issues with downloads.openwrt.org?
<rsalvaterra>
aparcar: Opened fine here. Any specific issue?
<aparcar>
I guess it was a local issue, sorry for the noise
Borromini has joined #openwrt-devel
<Borromini>
stintel: ping
dangole has joined #openwrt-devel
<[florian]>
aparcar: no news, and yes
<rsalvaterra>
I just did a git grep CONFIG_COMPAT_32BIT_TIME target/linux/ and I'm horrified.
<rsalvaterra>
Time to do some cleaning.
Gaspare has joined #openwrt-devel
<dwmw2_gone>
dangole: hi, any progress on the SATA thing?
<aparcar>
[florian]: all right don't hesitate to vote, in case you're in favor, it's a majority
rua has quit [Ping timeout: 480 seconds]
rua has joined #openwrt-devel
Gaspare has quit [Ping timeout: 480 seconds]
<[florian]>
aparcar: will do
<aparcar>
[florian]: thanks :)
<dangole>
dwmw2_gone: I've compared all relevant registers in the state after Linux has booted and I can't spot any (relevant) difference. maybe I'm not yet including the right registers in the dump... or what is missing may be hiding behind some indirect access or just requires initialization...
<dangole>
dwmw2_gone: so if you have any good ideas what to do next...
<dangole>
dwmw2_gone: maybe you can try TFTP-booting the vendor image using the new U-Boot and see if that works?
<dwmw2_gone>
Did you try booting one U-Boot from the other?
vchrizz1 has joined #openwrt-devel
vchrizz has quit [Read error: Connection reset by peer]
rua has quit [Remote host closed the connection]
<dwmw2_gone>
The vendor image ought to boot from our new U-Boot, I think?
sankanro has joined #openwrt-devel
vchrizz1 has quit [Remote host closed the connection]
danitool has quit [Quit: Cubum autem in duos cubos, aut quadratoquadratum in duos quadratoquadratos]
vchrizz has joined #openwrt-devel
rua has joined #openwrt-devel
sankanro has quit [Remote host closed the connection]
dedeckeh has quit [Remote host closed the connection]
<stintel>
man that UniFi Switch Flex is almost perfect
<stintel>
I mounted one on the bottom of my desk, powering a GS108Tv3, a Cisco C7941 IP phone, and a spare port so that I can play with a PoE-PD device without the need for an injector
mattytap_ has quit [Ping timeout: 480 seconds]
<stintel>
Borromini: do you still have stock firmware on your EAP615? :P
SamantazFox_ has quit [Read error: No route to host]
SamantazFox__ has joined #openwrt-devel
<Borromini>
stintel: it's on the way
<Borromini>
the EAP
<Borromini>
at this point i'm trying to resuscitate my XGS1250... I flashed some shoddy code Birger put together
<Borromini>
and the OpenWrt kernel panics... and that's the point where you find out effing ZyXEL locked their uboot :^)
<stintel>
ah you don't have it yet
<Borromini>
nope
<Borromini>
so out of my depth with this ch314a thingy
<Borromini>
giving me the evil eye
<stintel>
neggles: piong
rua has quit [Ping timeout: 480 seconds]
<neggles>
stintel: hello, yes, this is dog
<neggles>
ah, eap615 tree update?
<neggles>
borromini: what kind of ch341 trouble? they’re generally fairly solid these days
<stintel>
neggles: I think I'm ready with EAP615-Wall, yes. just haven't tested the GPIO buttons
<stintel>
but you have a US device, might not be compatible
<neggles>
IIRC they match what’s in the tp-link gpl tarball DTS
<Borromini>
but I probably did it wrong and should have passed the whole 'MX25L12835F/MX25L12845E/MX25L12865E' value?
<neggles>
“MX25L12835F/MX25L12845E/MX25L12865E"
<neggles>
yeah
<Borromini>
ok, thanks
<Borromini>
learning as we go :)
<neggles>
it’s not very intuitive, it’d be nice if you could supply any of those individual models
rua has joined #openwrt-devel
<Borromini>
yes, that's what confused me
<neggles>
took me a while to work that one out
<Borromini>
=)
<neggles>
I’ve got one of those $9 allwinner f1c200s boards I’ve been meaning to make into a dedicated flash programmer… though tbh I have about four or five unfinished spi flash programmer device designs
<Borromini>
that sounds neat
<neggles>
i wonder if they make 256mbit SPI PSRAM chips or if 128 is the max
<neggles>
though actually the allwinner would be able to handle what I’m thinking of - attach SPI flash chip, plug device into USB port, it shows up as a mass storage device with a file you can copy to read/dump the chip + a .txt with device info
<neggles>
we love adorable single-chip Linux device
<Borromini>
=)
<neggles>
it’s a 0.4mm pitch QFN88, ARM926EJ-S, has 64MB of built in RAM
<neggles>
hand-solderable with a little bit of care
<neggles>
(and a hotplate/hot air pencil)
<Borromini>
i have zero soldering skills and experience, unfortunately
<Borromini>
did/do plan on learning it though
<neggles>
hehe, i wouldn't really suggest DIYing a board just to play with
<neggles>
Borromini: at least, not when you can get an f1c200s board for very little https://www.aliexpress.com/item/1005003589764946.html (they're cheaper direct from the OEM but their shop is closed until the 8th of feb)
<Borromini>
:)
<Borromini>
i've dumped the flash contents, how should I go about modifying the uboot environment?
<neggles>
do you have the partition table? (also what device is this)
<Borromini>
XGS1250-10, and yes i do have a partition table
<neggles>
`strings` will extract the current environment from those without any effort, then to make a replacement you use `mkenvimage`
<neggles>
and some more dd trickery to merge them back into the dump
<Borromini>
ok, strings on the first dd indeed shows all the uboot-env settings nicely
<neggles>
the two env images should be identical
<neggles>
to run `mkenvimage` you will need a text file with all of the env settings in `key=value` form, which should be pretty much what you get from `strings`
<neggles>
does u-boot output to console & just not accept input? or is it totally silent?
<Borromini>
u-boot-env2.bin is empty
<Borromini>
it outputs to console
<neggles>
ha! classic
<neggles>
that just means it's not had the env changed since initial flash
<neggles>
or they're not actually using redundant environments
<Borromini>
yep full log here: u-boot-env2.bin iflag=skip_bytes,count_bytes skip=$((0xf0000)) count=6553
<Borromini>
no the u-boot-env2 partition is empty
<Borromini>
$ file u-boot*bin
<Borromini>
u-boot-env2.bin: ISO-8859 text, with very long lines, with no line terminators
<Borromini>
u-boot-env.bin: data
<neggles>
yeah, that's normal - if the default u-boot environment is still in use, i.e. it's not had `saveenv` run since it was flashed at the factory, env2 will be blan
<neggles>
default env is stored inside u-boot itself, don't need a third copy
<Borromini>
ah i see
<Borromini>
in that PR i was told to change the boot command to something bogus so that it falls back to a shell
<neggles>
you should be able to get into the console by powering it on and holding down spacebar on your serial terminal
<Borromini>
no no
<Borromini>
that does not work, i tried ... ZyXEL disabled it
<neggles>
odd.
<neggles>
hmm. pull the u-boot binary itself out - dd iflag=count_bytes if=flashdump.bin of=u-boot.bin count=$((0xe0000))
<neggles>
then run `strings -20` on that and you should find the default/baked-in env
<neggles>
but you could also set `bootcmd=help` or something like that
<Borromini>
maybe a silly question, but that would make it sit there and wait for input at every boot, no?
<Borromini>
(just wondering if i need to change it back to something sane once I get a functional firmware going again)
<Borromini>
i just ran strings -20, but it is spewing out a lot of stuff, no set env values though, even some html from what seems to be an embedded recovery
<Borromini>
oh i can write ZyXEL if you'd like apparently they are rather forthcoming
<Borromini>
will do :)
<neggles>
can't hurt to have
<neggles>
but i suspect the 1210 one will cover what i'm looking for
<Borromini>
at least the SoC is identical
<Borromini>
so i assume the platform is very similar
<stintel>
I can't get over the port layout on that zyxel :P
<Borromini>
request has been sent
<Borromini>
stintel: how so?
<Borromini>
neggles: sorry, i see my paste buffer fouled up the link to what should have been the full boot log, i can still link to it if you'd like but i suppose it won't help
<Borromini>
unlike on their plain gigabit switches OpenWrt supports now this one does not prompt for user input to interrupt booting on the console.
<neggles>
Borromini: it is fine :P
<neggles>
i suspect they've bypassed the usual autoboot
<neggles>
I don't really regret selling when i did though, crypto is a distributed pyramid scheme, someone else has to buy in for you to get paid out... and I got ~$400 out of what cost me maybe $10 in power at the time
<neggles>
yup that's the pupper!
<stintel>
haha this brings back memories
<neggles>
I think I mined more like 300k actually, i remember I sent 100k to the dogecoin nascar fund
<stintel>
I used to be mining LTC on p2pool, which announced found blocks in IRC